@colinlu50/openclaw-lark-stream 260327.2.4 → 260327.2.6

package/README.en.md

@@ -4,23 +4,28 @@ English | [中文](./README.md)
 
 ![demo](./demo.gif)
 
-Based on the official [openclaw-lark](https://github.com/larksuite/openclaw-lark) plugin, with **real-time streaming output**, **reasoning display**, and **tool call indicators**.
+Based on the official [openclaw-lark](https://github.com/larksuite/openclaw-lark) plugin, with **real-time streaming output** and **agent execution visibility**.
 
 ## ✨ What's Changed
 
 The official plugin delivers LLM block results all at once after completion. This version enables:
 
-- **Real-time streaming output** — each block's content is progressively appended to the streaming card as it's generated
+- **Real-time streaming output** — each block's content is progressively appended to the Lark card as it's generated
 - **Group chat streaming** — streaming output works in group chats as well
-- **Reasoning display** — think content from reasoning models (DeepSeek-R1, Claude 3.7, etc.) streams live
-- **Tool call indicators** — when the agent calls a tool, the card shows the current tool name in real-time
-- **Process panel** — on completion, all reasoning blocks and tool calls are folded into a single expandable "Process" panel in chronological order
-- **Token usage** — the card footer shows input/output token counts and context window usage percentage by default
+- **Agent execution visibility** — full transparency into the agent's reasoning and execution flow
+  - **Reasoning display** — think content from reasoning models (DeepSeek-R1, Claude 3.7, etc.) streams live
+  - **Tool call indicators** — when the agent calls a tool, the card shows the current tool name in real-time
+  - **Process panel** — on completion, all reasoning blocks and tool calls are collapsed into a single expandable panel in chronological order
+  - **Token usage** — the card footer shows input/output token counts and context window usage percentage by default
 
 ## 📢 News
 
-- **2026.3.27 v2** — Collapsed panels merged into a single "Process" panel; footer now shows token usage and context % by default
-- **2026.3.27 v1** — Compatible with OpenClaw >= 2026.3.22; added AskUserQuestion interactive tool; panels appear in chronological order; fixed card table limit error 230099
+- **2026.3.27**
+  - Compatible with OpenClaw >= 2026.3.22
+  - Added AskUserQuestion interactive tool
+  - Reasoning blocks and tool calls merged into a single expandable panel in chronological order
+  - Footer now shows token usage and context window percentage by default
+  - Fixed card table limit error 230099
 - **2026.3.23** — First release with real-time streaming output and tool call indicators (for OpenClaw < 2026.3.22, use the `0322` branch)
 
 ## 📦 Installation

package/README.md

@@ -4,23 +4,28 @@
 
 ![demo](./demo.gif)
 
-基于官方 [openclaw-lark](https://github.com/larksuite/openclaw-lark) 插件，支持**实时流式输出**、**推理过程展示**和**工具调用状态**。
+基于官方 [openclaw-lark](https://github.com/larksuite/openclaw-lark) 插件，支持**实时流式输出**和 **Agent 执行过程可视化**。
 
 ## ✨ 改动说明
 
 官方插件在 LLM 生成完一个 block 后才一次性推送结果。本版本实现了：
 
-- **实时流式输出** — 每个 block 的内容在生成过程中逐步追加到流式卡片
+- **实时流式输出** — 每个 block 的内容在生成过程中逐步追加到飞书卡片
 - **群聊流式输出** — 群聊中也可使用流式输出
-- **推理过程展示** — 推理模型（DeepSeek-R1、Claude 3.7 等）的 think 内容实时流出
-- **工具调用状态** — agent 调用工具时，卡片顶部实时显示当前工具名称
-- **思考过程面板** — 完成后，所有推理块和工具调用按发生顺序折叠进一个「思考过程」可展开面板
-- **Token 用量展示** — 卡片底部默认显示 input/output token 数和 context 使用百分比
+- **Agent 执行过程可视化** — 完整还原 agent 的推理与执行流程
+  - **推理过程展示** — 推理模型（DeepSeek-R1、Claude 3.7 等）的 think 内容实时流出
+  - **工具调用状态** — agent 调用工具时，卡片顶部实时显示当前工具名称
+  - **思考过程面板** — 完成后，所有推理块和工具调用按发生顺序折叠进一个可展开面板
+  - **Token 用量展示** — 卡片底部默认显示 input/output token 数和 context 使用百分比
 
 ## 📢 News
 
-- **2026.3.27 v2** — 折叠面板合并为单个「思考过程」；底栏默认显示 token 用量和 context 使用率
-- **2026.3.27 v1** — 适配 OpenClaw >= 2026.3.22；新增 AskUserQuestion 交互式提问工具；推理块与工具调用面板按发生顺序排列；修复卡片表格超限错误 230099
+- **2026.3.27**
+  - 适配 OpenClaw >= 2026.3.22
+  - 新增 AskUserQuestion 交互式提问工具
+  - 推理块与工具调用按发生顺序合并为单个可展开面板
+  - 底栏默认显示 token 用量和 context 使用百分比
+  - 修复卡片表格超限错误 230099
 - **2026.3.23** — 发布第一版，支持实时流式输出和工具调用状态展示（适配 OpenClaw < 2026.3.22，请切换到 `0322` 分支）
 
 ## 📦 安装

package/dist/index.js

@@ -114832,16 +114832,15 @@ function formatFooterRuntimeSegments(params) {
     }
   }
   if (footer?.context && metrics) {
-    const freshTotal = metrics.totalTokensFresh === false ? void 0 : metrics.totalTokens;
-    const total = typeof freshTotal === "number" ? Math.max(0, freshTotal) : void 0;
+    const used = typeof metrics.inputTokens === "number" ? Math.max(0, metrics.inputTokens) : void 0;
     const ctx = typeof metrics.contextTokens === "number" ? Math.max(0, metrics.contextTokens) : void 0;
-    if (total != null && ctx != null) {
-      const totalLabel = compactNumber(total);
-      const ctxLabel = compactNumber(ctx);
-      const pct = ctx > 0 ? Math.round(total / ctx * 100) : 0;
+    if (used != null && ctx != null) {
+      const pct = ctx > 0 ? Math.round(used / ctx * 100) : 0;
       if (verbose) {
-        zhParts.push(`\u4E0A\u4E0B\u6587 ${totalLabel}/${ctxLabel} (${pct}%)`);
-        enParts.push(`Context ${totalLabel}/${ctxLabel} (${pct}%)`);
+        const usedLabel = compactNumber(used);
+        const ctxLabel = compactNumber(ctx);
+        zhParts.push(`\u4E0A\u4E0B\u6587 ${usedLabel}/${ctxLabel} (${pct}%)`);
+        enParts.push(`Context ${usedLabel}/${ctxLabel} (${pct}%)`);
       } else {
         zhParts.push(`${pct}% ctx`);
         enParts.push(`${pct}% ctx`);
@@ -115923,6 +115922,7 @@ var init_streaming_card_controller = __esm({
         if (!this.shouldProceed("onToolStart")) return;
         const toolName = payload.name ?? "unknown";
         log17.debug("onToolStart", { toolName, phase: payload.phase });
+        if (payload.phase === "update") return;
         this.commitReasoningEvent();
         this.reasoning.isReasoningPhase = false;
         this.completeRunningTools();